Crystal Structure

In general, by solid we mean an equilibrium state with broken translation

symmetry. That is a state for which there exist observablessay, densities
of particleswith spatially dependent expectation values, A(r) 6= A(r0 ). A
crystal is a solid that features translation symmetry:
A(r) = A(r + T) ,

(1)

where T is one of the translation vectors. For a crystal, there exist an

infinite number of translation vectors, but in d dimensions the number of
linear independent translation vectors cannot be larger than d. For a genuine crystal, the number of linear independent translation vectors is exactly
equal to d. There exist so-called quasi-crystals for which the number of linear independent translation vectors is smaller than d. Solids that do not
possess translation symmetry at all are called amorphous solids, or structural glasses. In this course, we will be mostly dealing with genuine crystals.
The group of translations
All the translations with respect to which a given crystal is symmetric
in the sense of Eq. (1) form a group. A group G is a set of objects obeying
the following four axioms.
A1. For any two elements A and B of the group G there exists a group
operation (sometimes called a composition, a product, etc.) such that
A B G.
A2. Associativity: (A B) C = A (B C).
A3. There exists an identity element E G such that E A = A E = A.
A4. For each A G there exists an inverse element X G such that
X A = A X = E.
From these axioms there are two elementary uniqueness theorems: (i)
The identity element is unique, and (ii) For each A G its inverse element
is unique. It is also trivially seen that if X is inverse to A, then A is inverse
to X; and E is inverse to itself.
In the Abelian group, the group operation is commutative: A B = B A.
To make sure that all the translations that carry a crystal into itself
form a group, GT , we observe that the elements of GT can be associated
with the translation vectors. Then the composition of two translations with
the vectors T1 and T2 is yet another translation carrying a crystal into itself,
with the translation vector T3 = T1 + T2 . Since the addition of vectors is
associative, the axiom A2 is satisfied. The identity element is the translation
1

by zero vector. And for the translation by the vector T the inverse element
(trivially belonging to GT ) is the translation by the vector T. Since the
vector addition is commutative, the translation group is Abelian.
Primitive sets of translation vectors. Bravais lattice
Each translation vector T of the group GT can be parameterized as a
linear combination, with integer coefficients, of specially chosen three (for
definiteness we work in three dimensions) translation vectors (a1 , a2 , a3 ):
T Tn = n1 a1 + n2 a2 + n3 a3 .

(2)

Here we use a convenient integer-vector notation for the three integer coefficients: n = (n1 , n2 , n3 ). Corresponding set of the three vectors is called
primitive. The requirement that each of the numbers (n1 , n2 , n3 ) be integer significantly constrains the choice of primitive sets. For example if we
1 = 2a1 , the new set
take the primitive set (a1 , a2 , a3 ) and replace a1 a
(
a1 , a2 , a3 ) will not be primitive, because a1 = 0.5
a1 . Still, there is an infinite number of choices of a primitive sets. An elementary step that allows
one to go from one primitive set to yet another one is by replacing one of
the vectors in the set by a sum of itself and the other vector of the same
set, like a1 a1 + a2 , etc. Note that iteratively repeating this procedure
we can arrive at a primitive set having vectors of arbitrarily large lengths
(but at the expense of having arbitrarily small angles between them)!
In accordance with the parameterization (2), all the translation vectors
of the group GT form a lattice, each site of which is conveniently labelled
with the integer-vector subscript n. This lattice is called Bravais lattice
(after Auguste Bravais who introduced it in 1850). The geometric shape
of the Bravais lattice does not depend on the choice of the primitive set.
Indeed, an alternative way of introducing the Bravais lattice is to take some
point in spacecall it the originand consider the set of all points that can
be obtained from the origin by applying all the translations of the group GT .
Hence, the Bravais lattice is nothing but a a geometric way of visualizing
GT . What does depend on the choice of the primitive set, however, is the
labeling of the sites of the Barvais lattice. The integer vector n ascribed to a
given site of the Bravais lattice is unambiguously related to the choice of the
primitive set, and, vice versa, if each site of the Bravais lattice is naturally
labelled with an array of integers (n1 , n2 , n3 ), then corresponding primitive
set is given by the three translation vectors going from the origin to the sites
(1, 0, 0), (0, 1, 0), and (0, 0, 1).
2

a2

a1

Figure 1: Hexagonal lattice: |a1 | = |a2 |, the angle between a1 and a2 is

2/3. The vectors a1 and a2 are the primitive set forming the primitive
parallelogram. The dashed hexagon is the Wigner-Seitz primitive cell.

With the notion of Bravais lattice, it is easy to exhaustively describe

all possible choices of primitive sets. A translation vector T is primitive if
it cannot be represented as an integer multiple of yet another translation
vector T0 (that is it cannot be represented as T = mT0 , with some integer
m > 1). By definition, each of the vectors in a primitive set is primitive.
Does it work other way around: Can we construct a primitive set starting
from any primitive vector? The answer is yes, and we are in a position
to describe all possible ways of doing that. It is convenient to start with
2D, where a choice of a primitive translation vector a1 splits the Bravais
lattice into a stack of one-dimensional latticeslines of pointsthe vector
a1 being the primitive translation vector for each of them. Consider the line
that contains the origin of our 2D Bravais lattice, call it the baseline. There
are two lines that are the closest to the baseline. The two are equivalent,
so we pick up just one of them and call it the nearest line. It is rather
obvious now that any vector a2 that connects the origin to any point on the
nearest line can be chosen as a complimentary to a1 primitive vector to form
a primitive set (a1 , a2 ); and it is also obvious that there are no other choices
of a2 for given a1 . Now we go from 2D to 3D. Here one can start with any
plane of lattice points containing the origin and the fixed primitive vector
a1 . This plane (call it the base plane) is a 2D Bravais lattice for which we
can choosein accordance with the above-discussed 2D rulesa primitive
vector a2 such that (a1 , a2 ) is a primitive set. The original 3D lattice then
splits into a stack of 2D lattices obtained by translations of the base plane.
Take one of the two planes closest to the base plane, call it the nearest plane.
In a direct analogy with 2D case, all possible choices of a3 are then given
by vectors connecting the origin to one of the points in the nearest plane.
A natural choice of a primitive set, minimizing the lengths of the vectors
of the set, is as follows. Let a1 be the shortest (and thus obviously primitive)
translation vector. The vector a2 is then chosen to be the shortest of all
translation vectors carrying the origin to one of the Bravais lattice sites in
the line closest to the line of sites formed by translating the origin by integer
multiples of a1 . In 3D, the vector a3 is then chosen to be the shortest of
all translation vectors carrying the origin to one of the Bravais lattice sites
in the plane closest to the base plane formed by translating the origin by
n1 a1 + n2 a2 .
Primitive cell
A primitive cell is a geometric shape, consistent with the group GT , that
perfectly (i.e. without gaps and overlaps) tiles the whole space if translated
4

by the vectors of the group GT . There is a continuum of options of choosing

a primitive cell. A natural (and still infinite) class of primitive cells comes
with the parameterization (2), where each primitive set (a1 , a2 , a3 ) yields
a primitive parallelepiped (parallelogram in 2D) build on the vectors of the
given set. All primitive cells associated with the group GT have one and the
same volume Vc . Indeed, the primitive cell always contains exactly one point
of the Bravais lattice; otherwise there would be either gaps or overlaps when
tiling the space by translating the primitive cell with the vectors of the GT .
The invariance of Vc with respect to the choice of the primitive cell follows
then from a simple observation that the total number of the points of the
Bravais lattice in a certain large volume V  Vc is given by the ratio V /Vc
(with the accuracy controlled by the small parameter Vc /V  1). In other
words, Vc is nothing but the inverse density of the points of the Bravais
lattice.
With the parameterization (2), the explicit expression for Vc is given by
the mixed scalar product of the primitive translation vectors.
Vc = |a1 (a2 a3 )|

(d = 3) .

(3)

The 2D analog of Eq. (3) is

Vc = |a1 a2 | = |a1 ||a2 | sin

(d = 2) ,

(4)

where is the angle between the two primitive translation vectors. Equation
(3)/(4) is not only necessary, but also a sufficient condition for three/two
translation vectors to form a primitive set in 3D/2D. This is obvious from
the necessary and sufficient condition of having exactly one site of Bravais
lattice per primitive parallelepiped/parallelogram.
Along with primitive parallelepipeds/parallelograms, a popular choice
of the primitive cells is the Wigner-Seitz construction (which is a type of
Voronoi cell1 ), illustrated in K-Fig. 4 (here and below K-Fig. stands for
corresponding figure in the 8th edition of Kittel text). The Wigner-Seitz
primitive cell is especially convenient in the cases when crystal structure
features extra symmetries in addition to the translation group. The WignerSeitz primitive cell then readily reflects those symmetries, while the primitive
parallelepipeds/parallelograms tend to hide them. For an illustration, see
Fig. 1.
1

For any discrete set of point sitesnot necessarily forming a regular (crystal)
structurea Voronoi cell contains exactly one site, along with all the points in the
plane/space for which the given site is closer than any of the rest of the sites.

Figure 2: Honeycomb lattice. The bonds are to guide an eye. The lattice
sites correspond to the vertices of hexagons tiling the plane. The lattice is
not inversion-symmetric with respect to the lattice points, and thus is not
a Bravais lattice.

a1

a2

Figure 3: Using two colors, we see that the honeycomb lattice is a bipartite lattice formed by superimposing two identical hexagonal lattices, each
of which can be taken as a Bravais lattice of the honecomb lattice. This
allows one to readily construct a primitive set of translation vectors by considering one of the two hexagonal lattices. The dashed parallelogram shows
an alternative choice of the Bravais lattice and, correspondingly alternative primitive cell. In this case, Bravais lattice consists of the centers of
the hexagons. With this choice, the symmetry between the two hexagonal
sub-lattices is clearly seen.

Basis of crystal structure. Non-Bravais lattices

The primitive cell can be viewed as a set of potentially non-equivalent
spatial points forming the crystal structure. The word potentially expresses
the reservation that in addition to the translation symmetry a crystal structure can feature some extra symmetries rendering different points within a
primitive cell equivalent. The crystal structure is then unambiguously fixed
by specifying the values of relevant functions (say, density distributions)
within the primitive cell. A primitive cell with specified functions within it
is called the basis of crystal structure.
In many cases we are interested in some special points within the crystal structure, say, the maxima of the density distributions (these maxima
are often loosely referred to as positions of corresponding atoms). A crystal structure of discrete points is called crystal lattice, or just a lattice. A
Bravais lattice is the simplest example of a crystal lattice. It is a minimal
crystal lattice corresponding to a given group GT . Clearly, not any lattice is
a Bravais lattice. For a lattice to be a Bravais lattice it is necessary and sufficient to have exactly one lattice point per primitive cell. Correspondingly,
if the primitive cell contains two or more lattice points, it is a non-Bravais
lattice.
In the next sub-section, we will be discussing extra symmetries of crystal
structure. Here we introduce one of them which proves extremely helpful for
identifying non-Bravais lattices of identical points. The inversion symmetry
is the symmetry with respect to changing the sign of all the coordinates
with respect to a certain origin (called the center of inversion). All the
Bravais lattices are inversion-symmetric with respect to inversions against
the lattice points, because for any translation vector T, the vector T is
also a translation vector (the group property). Hence, if a lattice is not
inversion-symmetric with respect to each of its points, it is guaranteed to be
a non-Bravai lattice!
An important2 example of non-Bravais lattice is the honeycomb lattice,
see Figs. 2 and 3.
Extra symmetries of crystal structure
In addition to the translation group GT , there can exist other symmetry
2

And extremely popular nowadays because of graphene (Nobel Prize 2010).

operations that carry a given crystal structure into itself.3 The most typical
of those operations is a rotation about an axis. The crystallographic restriction theorem states that only a few special rotation angles (modulo 2) are
consistent with discrete translation symmetry: 2/2, 2/3, 2/4, and 2/6.
Corresponding axes are conveniently denoted by the symbols 2, 3, 4, and
6. The proof of the theorem is very simple in the case when the rotation
axis is perpendicular to one of the translation vectors (which is automatic
in 2D). This is also the case for any rotation that carries the Bravais lattice
into itself, since it implies the existence of two sites of Bravais lattice in a
line perpendicular to the rotation axis.
Let R be the operator of the rotation by the angle that carries the
crystal structure into itself. Without loss of generality we assume that 6= ,
since were are not going to rule out the case = . The existence of
the translation vector perpendicular to the rotation axis then guaranties
the existence of a plane of the sites of Bravais lattice perpendicular to the
rotation axis. Let a be the shortest translation vector in this plane. Clearly,
the pair (a, Ra) of shortest translation vectors forms a primitive set in this
plane, and the vector R2 a can be represented as
R2 a = nRa + ma ,

(5)

where n and m are some integers. [At this point one can notice that Eq. (5)
applies also to the case = , since in this case Ra = a.]
Writing (5) in Cartesian componentswith the x-axis along a and the
z-axis being the axis of rotationwe get
"

cos 2 sin 2
sin 2 cos 2

#"

1
0

"

=n

cos sin
sin cos

#"

1
0

"

+m

1
0

(6)

which yields a system of two constraints:

(

(2 cos n) cos = m + 1 ,
2 sin cos = n sin .

(7)

From the second constraint we see that either sin = 0 or cos = n/2 (in
both cases the first constraint is then easily met by an appropriate choice of
m). This proves the statement. (The proof for a general case can be found
at http://en.wikipedia.org/wiki/Crystallographic restriction theorem.)
3

The symmetry of the crystal structure can be naturally lower than the symmetry of
corresponding Bravais lattice, since Bravais lattice is the minimal possible crystal structure
with corresponding translation group.

The group of rotations about a given l-fold axis (l = 2, 3, 4, 6) is an

example of a point group. The crystallographic point group or crystal class
is the mathematical group comprising the symmetry operations that carry
a crystal into itself while leaving at least one point unmoved. The other
point symmetry operations are: reflection, which reflects the structure across
a reflection plane; inversion, which changes the sign of the coordinate of
each point with respect to a center of symmetry or inversion point (in 2D,
inversion is the same as the angle- rotation, and the inversion symmetry is
thus the same as the 2-fold axis with the same center); improper rotation,
which consists of a rotation about an axis followed by an inversion. Rotation
axes (proper and improper), reflection planes, and centers of symmetry are
collectively called symmetry elements. In 3D, there are 32 possible crystal
classes. Each one can be classified into one of the seven crystal systems.
The space group of the crystal structure is composed of the translational
symmetry operations in addition to the operations of the point group. Apart
of the group GT of pure translation which each crystal structure has by
definition, space groups can include screw axes, which rotate a point around
an axis while translating parallel to the axis and glide planes, which reflect a
point through a plane while translating it parallel to the plane. In 3D, there
are 230 distinct space groups. The 2D analogs of space groups are called
wallpaper groups (or plane crystallographic groups); there are 17 distinct
groups in 2D.
Sometimes one is interested in the symmetry of the Bravais lattice only.
Since not any crystal lattice is a Bravais lattice, we can expect that the
number of distinct symmetry groups of Bravais lattices is smaller than the
total number of crystallographic groups. And this is indeed the case. In 2D
there are only 5 distinct symmetry groups (often referred to as symmetry
types) of Bravais lattices. That is only 5 out of 17 wallpaper groups are
relevant to the symmetries of 2D Bravais lattices. One of the five lattice
types is generic, it is referred two as oblique (a.k.a parallelogram) Bravais
lattice. The other foursquare, hexagonal, rectangular, and centered rectangular (a.k.a rhombic, or isosceles triangular), see K-Fig. 7are special,
they feature extra symmetries.
The square lattice (symmetry group p4m) has two rotation centres of
order 4, and reflections in 4 distinct directions (horizontal, vertical, and
diagonals).
The hexagonal lattice (symmetry group p6m) has one rotation centre of
order 6, two rotation centres of order 3, and three of order 2. It has also
reflections in 6 distinct directions.
The rectangular lattice (symmetry group pmm) has reflections in two
10

perpendicular directions (parallel or perpendicular to the sides of the primitive rectangle), and four rotation centres of order 2 located at the intersections of the reflection axes.
The centered rectangular (a.k.a rhombic, or isosceles triangular) lattice
(symmetry group cmm) has reflections in two perpendicular directions (parallel to diagonals of the primitive rhombus), and a rotation of order 2 whose
centre is not on a reflection axis. It also has two order-2 rotations whose
centres are on a reflection axis.
In 3D there are only 14 symmetry types of Bravais lattices (only 14
out of 230 space crystallographic groups are relevant to the 3D Bravais
lattices), see Fig. 5. The triclinic is the generic 3D type, an analog of
of the oblique lattice in 2D, the other 13 feature some extra symmetries.
In 3D, along with lattice types there are also lattices systems. The idea
of systems is to group the lattice types sharing certain similarity. There
are 7 Bravais lattice systems: triclinic, monoclinic, orthorombic, tetragonal,
trigonal, cubic, hexagonal. Three out of the 7 systemstriclinic, trigonal,
and hexagonalare just synonyms to corresponding Bravais lattice. The
other four systems contain two and more Bravais lattices each. A very
simple and important cubic system contains three Bravais lattices: simple
cubic (sc), body-centered cubic (bcc), and face-centered cubic (fcc), see Fig. 5.
Close-packed structures
An important class of lattices is formed by close-packed identical spheres.
A macroscopically large system of identical spheres is called close-packed if
it maximizes the packing fraction defined as the ratio of the volume occupied
by the spheres to the total volume of
the system. The statement that the
maximal packing fraction equals /(3 2) 0.74048 has been know for a
long time as Kepler conjecture. In 1998 Thomas Hales, following an approach
suggested by Fejes T
oth (1953), announced his proof of the Kepler conjecture
(by exhaustion involving checking of many individual cases using complex
computer calculations). Hales proof is very close to being accepted as a
theorem.
There are infinitely many (!) ways to achieve the Keplers packing fraction. One starts with a close-packed layer (2D hexagonal lattice ) of identical
spheres, see K-Fig. 19, and observes that there are two equivalent ways of
putting the next layer on top of the previous one. Correspondingly, a closepacked 3D structure of the layers can be described as a periodic, or nonperiodic sequence of three letters, A, B, and C, characterizing three distinct
types of the layers. (The only restriction on the sequence is that two ad11

a1
a3
a4

a1
a2

a2

Figure 4: The centered rectangular (a.k.a rhombic, or isosceles triangular)

lattice. Solid lines show the reflection axes. The following three constraints
have to be satisfied: (i) |a4 | =
6 |a2 | , (ii) |a1 | =
6 |a2 | , and (iii) |a1 | =
6 |a4 |.
Indeed, if |a4 | = |a2 |, the lattice becomes a square lattice, and if either
|a1 | = |a2 | or |a1 | = |a4 |, the lattice is a hexagonal lattice. The rectangular
cell built on the vectors a2 and a4 is not primitiveit contains two sites.
This cell gives the main name to the lattice. This cell is very convenient
for visualizing all the symmetries of the centered rectangular lattice. The
set (a1 , a3 ) defines the primitive rhombus (|a3 | = |a1 |), explaining why the
lattice is also called rhombic. The primitive parallelogram build on the set
(a1 , a2 ) consists of two isosceles triangles explaining the origin of yet another
name of the lattice.

12

triclinic

monoclinic

orthorhombic

tetragonal

sc

bcc

fcc

cubic

trigonal

hexagonal

Figure 5: The 14 types of 3D Bravais lattices grouped into 7 systems.

13

C1

A3

A2

A1

A2

A1

B
A6

A3
A4

A4

A6

A5
C2

C3
A5

Figure 6: Revealing the ABCABC . . . close-packing layers in the fcc lattie.

Shown are only the three visible faces of the cube. The diameter of the
close-packed spheres equals |A1 A2 |. The spheres centered at the points
A1 , A2 , . . . , A6 belong to a close-packed A-layer. The point B belongs to
a B-layer. A straightforwardbut not the shortestway to see that the
points C1 , C2 , C3 belong to a C-type layer (rather than a B-type layer) is
to consider the cubic units adjacent to the shown one. A more short and
elegant way is to invoke the inversion symmetry with respect to one of the
A-points. For example, the inversion with respect to the point A2 swaps
the points B and C1 , proving that the two points belong to the layers of
different types.

14

jacent letters be different.) The two most simple and important cases are
ABAB . . . and ABCABC . . . . The case ABAB . . . yields the hexagonal
close-packed (hcp) lattice. The hcp lattice consists of two hexagonal sublattices corresponding to the sets of A and B layers, respectively. Hence,
the Bravais lattice of the hcp lattice is hexagonal, and there are two points
in the primitive cell: an A-type one and a B-type one. The ABCABC . . .
packing corresponds to the fcc lattice, see Fig. 6. It is instructive to note
that the ABAB . . . packingas opposed to the ABCABC . . . packing
features reflectional symmetry with respect to the planes of the A and B
layers. On the other hand, the ABCABC . . . packingas opposed to the
ABAB . . . packingfeatures inversion symmetry with respect to the centers of the spheres; the inversion with respect to an A-center swaps B and
C layers, etc. It is worth recalling that the mere fact that the ABAB . . .
packing does not have inversion symmetry with respect to the lattice points
implies that corresponding hcp lattice is not a Bravais lattice.
Closed-packed lattices are typical, e.g., for noble gases. Apart from
helium, all solid phases of the noble gases have the fcc lattice. The solid
4 He has the hcp lattice, except for a small and essentially finite-temperature
region of the bcc lattice, see Fig. 7.
Sodium chloride and cesium chloride structures
Sodium chloride and cesium chloride give the names to two very common
two-color cubic crystal structures shown in Figs. 8 and 9. Note that in both
cases applying the colors changes the type of the Bravais lattice (within the
same cubic system).

15

Figure 7: Phase diagram of 4 He.

16

Figure 8: Sodium chloride structure. This two-color structure is obtained by

alternating coloring the points of a simple cubic lattice. The Bravais lattice
of the sodium chloride structure is fcc. The basis of the structure contains
two points of different colors.

17

Figure 9: Cesium chloride structure. This two-color structure consists of two

simple cubic sub-lattices of different colors. The two sub-lattices are translated with respect to each other by the vector (1/2, 1/2, 1/2). The Bravais
lattice of the cesium chloride structure is sc. The basis of the structure
contains two points of different colors.

18